Several design approaches are available to deal with flexibility in robotic systems. The brute-force approach is to make the robot's structural and drive members more massive and rigid. But in this approach the resulting large inertial forces require large actuators and the cost becomes high. Instead, system integrators are installing motor controllers to cancel excited vibrations in the robot.
Researchers at CAMotion Inc. (Atlanta, GA) have developed an optimized filter to fix these vibrations. However, their approach requires prior knowledge of the system's natural frequencies. These frequencies must be measured to be adjusted as the robot's arm moves to different places in the work space or to impose longer operational delays to reduce sensitivity to natural frequency variations. By coupling this command filtering with learning control on repetitive motions, the errors between commanded and achieved positions are reduced.
Developed by Steve Dickerson, founder of Dickerson Vision Technology, now DVT Corp. (Norcross, GA), CAMotion has embedded its motion-control algorithms in a series of scalable computer-controlled x-y-z gantry systems. To locally control the axes of these robotic systems, the company has built a remote-axis serial-interface device (RASID) that consists of a DSP-based motor controller with supervisory control provided by a PC. In operation, the RASID receives reference trajectories and learned feed-forward motor currents from the PC via a USB interface. It then performs proportional-plus-derivative control of linear or rotary permanent magnet drives.
CAMotion has installed a motion-control system at Vulcan Electro-Coating (Bessemer, AL), a company that provides both cathodic electrodeposition coating and powder coating of metallic parts for the automotive industry. The CAMotion system inspects the paint quality of automotive parts and then packs the parts in cardboard shipping containers. Parts inspection is performed in less than five seconds per part with the acquisition and analysis of as many as 24 images of each part. The packing machine can handle more than 500 parts per container and can load parts every four seconds. According to Dickerson, defects as small as 0.02 in. can be detected on the 36-in.-long parts.